When the music changes so does the dance – speakers’ insights at XTM LIVEStream
The shift to online has forced the hand of many companies in how they do business. The Intelligence Revolution has also impacted how companies approach globalization. How can you efficiently navigate the language landscape in this challenging business climate? On December 16th you will hear from localization experts on a variety of topics covering Intelligence Revolution, MT, AI, automation, and other developments in the localization industry.
This is another pre-event chat in a series of talks leading up to XTM LIVEStream – in the previous post we presented speakers’ take on language technology, AI, automation and the cloud. This time, we asked a few of the panelists – Ken Behan (SYSTRAN Software), Konstantin Savenkov (Intento, Inc.), and Andrzej Zydroń (XTM International) to share their thoughts on AI, the Intelligence Revolution, and how this shift has impacted them.
Ken Behan is VP of Sales & Marketing for SYSTRAN Software Inc. responsible for defining and implementing the global growth strategy. With 22 years’ experience in the language intelligence industry, he’s considered a thought leader in Natural Language Processing having held several Senior Executive roles within the industry.
We asked Ken Behan (SYSTRAN Software) how AI can help solve the challenges in the industry. He pointed out that while the pandemic has undermined our ability to communicate face to face, it has given us new ways to interact “as more people accept that videoconferencing is an acceptable way to communicate. AI can better support summation of such calls, analyzing what has been said and providing recommended solutions, and hopefully, in the near future do without the notion of a language barrier.”
Jean Senellart (CEO at SYSTRAN Software) recently noted that NMT technological innovation has reached a dead end and is no longer disruptive on its own. Ken’s view is that NMT improvements in terms of quality will be incremental “with the emphasis on shifting to domain specialization and constant engine training. In the next 5 years, MT innovation will move towards “audio and seamless conversations in any language. High-quality sentence parsing is one of the bigger challenges that will have to be solved so that conversations will seem natural.”
The demand for content to be available in their native language has moved from being a differentiator to a business necessity.
The shift to digital had a profound impact on how companies view localization. As more people spend more time online “the demand for content to be available in their native language has moved from being a differentiator to a business necessity. I’ve been surprised with how large corporations have suddenly recognized the importance in communicating with their clients in their native language,” explained Ken.
Asked if the pandemic had any impact on Ken’s routine and digital habits, he told us that “The most surprising part of the shift to work from home has been the ease that it has been, so in many ways my work practices haven’t changed. I did embark on one COVID-19 project which was to read Ulysses by James Joyce, which I completed in 3 months. Even as a native Dubliner, I’m still none the wiser as to what it’s about but at least I can now say I’ve read this classic novel!”
Konstantin Savenkov is CEO of Intento, Inc. After earning a Ph.D. in 2008, he led research and development efforts for online content services, then worked as CTO at Zvooq and as a chief operating officer at Bookmate. In 2016, he contributed his experience in artificial intelligence (AI), tech and operations to launch Intento, Inc., where they build tools to discover, evaluate and integrate best-of-breed MT and other cognitive AI services.
We asked Konstantin Savenkov (Intento, Inc.), about how AI can help solve the greatest challenge(s) for enterprises. He explained why for the bulk of companies localization is a complex venture to undertake; “For many enterprise localization scenarios, human labor cost is prohibitive. This is why a typical enterprise translates about 5-10% of content they would like to be translated.” By applying AI from day one “enterprises can dramatically increase the volumes of translated content they can manage, and as a result, they discover new audiences, and new markets. They can go global while being local from day 0, while without an AI approach, companies often only focus on important markets and products.”
He also doesn’t agree that NMT has reached a plateau, a point at which no further innovation is viable. Konstantin went on to explain “We see that after the release of the first public NMT (Google Translate) it was not immediately picked up by enterprises. The enterprise adoption grew in 2018-2019 when domain adaptation and custom terminology support were introduced and became more and more available.” According to Konstantin, there are still some obstacles on the path to full adoption, “We still see that many scenarios are blocked by the inability to control the tone of voice or gender bias, and by the lack of context-aware custom terminology support. Next to go will be translation style control and full document translation, as well as the ability to train on small amounts of data. Those developments will l lead to explosive growth of the enterprise MT adoption in the following years.”
We expect to see more effort going into improving MT beyond the objective linguistic quality to preferential requirements, such as style, tone of voice, gender, and others. We already do quite a bit of work in this direction at Intento and expect some support from MT vendors as well.
When it comes to machine translation and where it’s heading, Konstantin said, “We expect to see a much better MT quality for low-resource language pairs and domains, thanks to research in massively multilingual MT. Just a week ago, Facebook published its M2M-100 model, and we’ll see more of that in 2021.” He also points to full-text MT adoption, “Some first steps have been taken in full-text machine translation, providing consistent translations of multi-sentence texts. The light-weight interactive adaptation that today you may find in Lilt, ModernMT, and Systran, is expected to get wider support.” In Konstantin’s view, there will be more focus on enhancing MT on tone and voice aspects, “We expect to see more effort going into improving MT beyond the objective linguistic quality to preferential requirements, such as style, tone of voice, gender, and others. We already do quite a bit of work in this direction at Intento and expect some support from MT vendors as well.”
Another interesting trend to observe are MT providers that offer services to assess the quality of the MT output, “And another hot topic is a different kind of models, which estimate the MT quality on the segment level in real-time, without reference translations. For now, we see a number of emerging vendors in this space, and also a number of very different use-cases for this technology. I would estimate it will take a year or two to converge and provide something that will be accepted by the market,” noted Konstantin.
Has the migration to digital had any impact on MT buyers? As the content volumes for translation increased rapidly “we see MT owners in our customer organizations being approached by more and more other departments who need to translate their content.”
Asked if he attended any online conferences, Konstantin said ”I was able to tune in AMTA and Localize 2020 talks and found those extremely educating, especially regarding how different enterprises onboard MT and realize their ROI. For some reason, I found that at online conferences people are more willing to share interesting insights than at offline events. Such conferences feel more casual, and that’s a great thing.”
Andrzej Zydroń is one of the leading IT experts on Localization and related Open Standards. Following a career creating localization systems for large enterprises, Andrzej was a co-founder of XTM International and now, as CTO, oversees the technical architecture and product strategy with a strong emphasis on usability and the security of XTM.
Also contributing to the discussion was Andrzej Zydroń (CTO at XTM International). He shared with us his observations on key challenges that AI needs to solve and what’s next for NMT, “There are still a few challenges concerning the treatment of inline elements, the enforcement of correct terminology and how to cope efficiently with new words that were not in the original training material. The key driver for NMT at present is the quality and scope of the training material”.
Although the pandemic will have a tendency to reduce the speed of globalization, the human instinct for communication will continue to drive the need for technology to plug the gap between supply and demand in the translation industry.
The pandemic has accelerated the digital transformation and caused globalization to slow down. On a more positive note, however, Andrzej concludes, “The need for efficient translation management systems continues unabated. Although the pandemic will have a tendency to reduce the speed of globalization, the human instinct for communication will continue to drive the need for technology to plug the gap between supply and demand in the translation industry.”
Would you like to find out more about the Intelligence Revolution and how global companies turn it into an opportunity? Tune in to hear from Ken, Konstantin, and Andrzej at XTM LIVEStream on December 16th. Save your seat now!